TL;DR — Key Takeaways
-
OpenAI diversifies its chip strategy. A reported plan to purchase Google’s TPUs signals efforts to secure more AI computing capacity beyond NVIDIA.
-
Google strengthens its silicon ambitions. Expanded TPU availability positions Alphabet as an increasingly significant competitor in the AI hardware market.
-
Custom chips add another dimension. OpenAI’s Jalapeno processor project with Broadcom could give the company greater control over future AI infrastructure.
OpenAI plans to purchase artificial intelligence (AI) chips from rival Alphabet Inc.’s Google, marking the latest expansion in the ChatGPT maker’s aggressive campaign to secure critical computing power.
The prospective deal highlights growing commercial interdependence between the two tech giants, which compete fiercely in consumer AI and search products, according to The Information, citing people familiar with the matter.
Final financial terms remain undisclosed as the discussions are private. OpenAI and Google declined to comment.
The move comes amid an unprecedented infrastructure race across Big Tech.
Amazon.com Inc., Microsoft Corp., and Alphabet are projected to invest up to $600 billion combined this year in AI infrastructure, despite mounting scrutiny from Wall Street over when these historic capital expenditures will yield substantial returns.
OpenAI remains one of the largest buyers of NVIDIA Corp.’s market-leading graphics processing units (GPUs). However, surging demand and steep hardware costs have driven AI developers to pursue diversified sourcing strategies.
OpenAI previously began renting Google chips, and last week announced plans to deploy its frontier models through Google Cloud.
Google’s Tensor Processing Units (TPUs), first developed over a decade ago to accelerate internal search engine workloads, have emerged as the primary rival to NVIDIA’s silicon dominance. Originally reserved for in-house infrastructure, Google recently began selling TPUs directly to select enterprise customers and expanding access via third-party partnerships, including an agreement with Oracle Corp. to offer TPUs across its cloud infrastructure.
“The next generation of AI will require a diverse range of compute,” Amin Vahdat, a Google vice president, said in a statement accompanying the Oracle announcement.
Alphabet Chief Financial Officer Anat Ashkenazi has previously said the company expects to recognize initial revenue from external TPU sales agreements by year-end, with the vast majority anticipated to hit balance sheets in 2027. Commercial deals with leading AI labs, including rival Anthropic, have fueled steady enterprise demand for Google’s silicon as Alphabet prepares to report quarterly earnings.
For OpenAI, purchasing Google hardware forms part of a broader, multipronged silicon roadmap designed to eliminate single-vendor bottlenecks.
In parallel with external acquisitions, the startup has pushed into custom silicon development.
OpenAI’s in-house processor program, dubbed Jalapeno, is being developed in collaboration with semiconductor giant Broadcom Inc.
The initiative is spearheaded by Richard Ho, a former Google engineering leader who helped architect the search giant’s early TPU iterations. OpenAI plans to deploy its first custom-designed processors late this year as part of a multi-generation hardware roadmap.
